[spambayes-dev] What I'm working on...

Graham Toal gtoal at gtoal.com
Mon Oct 6 19:52:09 EDT 2003

A quick note to introduce myself and let you know what I'm working on.

I've hacked up an SMTP filter which you can put on an external system
in front of a proprietary mailer (Exchange, VMS, anything non-unixy that's
hard to tweak yourself) which filters SMTP sessions on the fly and allows
you to tag mail as it passes it through to a regular mail server.

It's crude but effective.  (You can see the WIP here:
http://www.gtoal.com/spam/smtpfilter.c.html  )

What I've discovered from experiments over the last 6 months or so is
that it *is* possible on a large scale to use a single Bayesian filter
on a large community as a group rather than having multiple training
databases for each user.  (Hope I'm not starting a flame war by admitting
that I've been using "spamprobe" rather than "spambayes" - I hadn't
discovered spambayes at the time I started this.  Spambayes does appear
to have one critical feature which makes it better, which is 'yes/no/maybe'
scoring rather than 'yes/no'...)

Anyway, the conclusion I have come to is to use a central filter on
the large group (and in my case the database is now 500Mb and *very*
well trained), but to allow the users to use a personal filter on their
PCs to correct any errors that they consider the group filter to have
made (eg if they *like* the crap that's sent out in "flowgo" mailing
lists...) - and that's why I'm looking at SpamBayes - it's the only
one I've found with an Outlook plugin and (relatively) easy to use
Windows installer.  The only thing I don't yet know is whether I can
set it up in a way that'll take stuff *out* of the spam folder and
put it back in the Inbox as a result of training.  I'm hoping that this
will be a trivial tweak!



More information about the spambayes-dev mailing list